Processing Doubly-Embedded Head-Final Relative Clauses Chien-Jer Charles LIN (National Taiwan Normal University) Thomas G. BEVER (University of Arizona) I. INTRODUCTION Theoretically, there can be 3 kinds of filler-gap relations in Doubly embedded head-final relative clauses (RCs): (i) Gap 1 ----- Filler 1 ----- Gap 2 ----- Filler 2 (ii) Gap 1 ----- Gap 2 ----- Filler 2 ----- Filler 1 (iii) Gap 1 ------ Gap 2 ----- Filler 1 ------ Filler 2 SERIAL DEPENDENCY

NESTED DEPENDENCY

CROSSING DEPENDENCY (not attested in human languages)

We focus on these two dependencies in Mandarin relative clauses.

Using doubly-embedded head-final relative clauses (RCs) in Mandarin Chinese, we examined the following effects in this study: • The effect of dependency types (Are serial dependencies easier than nested dependencies? YES – Bever 1970, Gibson 1998, among others) • The effect of extraction types (Are subject extractions easier than object extractions in Mandarin? YES, Lin & Bever 2006, Hsu et al., 2005; NO, Hsiao & Gibson 2003) A controversy in Mandarin relative clause processing: IS SUBJECT OR OBJECT RELATIVE CLAUSE EASIER TO PROCESS? In Lin & Bever (2006), we pointed out that Hsiao and Gibson’s (2003) study was confounded by DEPENDENCY TYPES across the conditions. They compared only between doubly-embedded subject RCs (a) and doubly-embedded object RCs (d), and found double object RCs read faster. This was actually an effect of serial dependencies (double ORCs) being easier than nested dependencies (double SRCs). It says nothing about the extraction effects.

Lin & Bever (2006):

(a) Subject RC embedded in Subject RC (Nested-SRC) GAP1

V1 GAP2 V2 N2 DE2 Head_Noun2 DE1 Head_Noun1 zhuisui xiahuai liumang de qiangfei de qiangfei follow scare rascal Rel robber Rel robber ‘the criminal that followed the robber that scared the rascals’

(b) Object RC embedded in Subject RC (Nested-ORC)

That double ORCs appeared easier in Hsiao & Gibson (2003) was not because object extractions were easier in Mandarin, but because they involved serial dependencies.

These doublyembedded Subject RCs in Hsiao & Gibson (2003) involved NESTED filler-gap relations.

MATERIALS Therefore, we found it necessary to examine the processing of doubly-embedded Mandarin RCs in a full 2x2 factorial design: Dependency type (serial vs. nested dep) X Extraction type (subject vs. object RC in the lower RC)

GAP1

V1 N2 V2 GAP2 DE2 Head_Noun2 DE1 Head_Noun1 zhuisui liumang xiahuai de qiangfei de qiangfei follow scare rascal Rel robber Rel robber ‘the criminal that followed the robber that the rascals scared’

(c) Subject RC embedded in Object RC (Serial-SRC) V1 GAP1 DE1 Head_Noun1 GAP2 V2 N2 DE2 Head_Noun2 zhuisui de qiangfei xiahuai liumang de qiangfei scare rascal Rel robber follow Rel robber ‘the criminal that the robber that scared the rascals followed’

(d) Object RC embedded in Object RC (Serial-ORC) V1 GAP1 DE1 Head_Noun1 zhuisui de qiangfei follow Rel robber ‘the criminal that the robber that the rascals scared followed’

N2 V2 GAP2 DE2 Head_Noun2 liumang xiahuai de qiangfei scare rascal Rel robber

II. METHOD

We collected self-paced reading data from 48 Mandarin speakers on 24 sets of experimental sentences. To avoid garden-path effect in reading the sentences, the data reported below were based on an experiment in which we specifically told the participants that all sentences contained RCs. Participants were also instructed the position of the RCs in the sentences (i.e. whether they modified the subjects or the objects of the main clauses).

III. RESULTS -- Comprehension accuracies and reading times on the lower embedded relativizers and the head nouns were compared across conditions. RTs of the doubly embedded Relativizer (DE2)

RTs of the lower RC regions (the shaded regions) Nested-SRC

15.96)

Nested-ORC

Serial-SRC

Serial-ORC

2000

Comprehension Accuracy

1800

1600

80%

1200

3500

1000

3000 2500

800 msec

100%

RTs of the doubly embedded Relativizer plus Head Noun (DE2 + HN2)

SRC ORC

600 400

msec

SRC ORC

40%

200 1200

2000

SRC ORC

1500 1000

1400

60%

msec

* Dependency effect: Serial dependencies > Nested dependencies (p < 0.001, F1 (1, 47) = 25.81; p < 0.01, F2 (1, 23) =

500

0

0

Nested

Serial

Nested

Serial

1000

20%

800

0% Nested

Serial

600 V/N

Comprehension Accuracy (%)

N/V

DE

Head N

* Dependency effect: Serial dependencies < Nested dependencies (p < 0.01, F1 (1, 47) = 11.92; p < 0.01, F2 (1, 23) = 9.25)

* Dependency effect: Serial dependencies < Nested dependencies (p < 0.001, F1 (1, 47) = 27.58; p < 0.001, F2 (1, 23) = 22.94)

* Extraction effect: Subject RCs < Object RCs (p < 0.05, F1 (1, 47) = 7.11; p < 0.05, F2 (1, 23) = 4.70)

* Extraction X Dependency: Subject RCs < Object RCs in Nested RCs (p < 0.05, F1 (1, 47) = 6.19; p < 0.05, F2 (1, 23) = 4.56)

Reading-Time Results (msec)

IV. DISCUSSION & CONCLUSION In the processing of doubly-embedded RCs in Mandarin, our data showed that the effect of dependency type was robust. Nested filler-gap relations were constantly harder than serial filler-gap relations across the board (in both comprehension accuracy & various RT data). Regarding the extraction effect, we found that when RCs were embedded in a nested filler-gap relationship (namely, when embedded in Subject RCs--conditions (a) & (b)), the extraction effect, with Subject-extracted RCs being read faster than Object-extracted RCs, was observed. In summary, our study demonstrated the following effects:

The effect of dependency types: Serial dependencies are easier than nested dependencies. (Bever 1970, Gibson 1998) The effect of extraction types: Subject extractions are easier than object extractions in Mandarin Chinese. (replicating and refining the results of Lin & Bever 2006, thus supporting a structure-based accessibility account; countering the claims of Hsiao & Gibson 2003 and their locality-based account) Selected References Bever, T. G. (1970). Cognitive basis for linguistic structures. In J. R. Hayes (Ed.), Cognition and the development of language (pp. 279-362). New York: Wiley. Gibson, E. (1998). Linguistic complexity: Locality of syntactic dependencies. Cognition, 68, 1-76. Hsiao, F., & Gibson, E. (2003). Processing Relative Clauses in Chinese. Cognition, 90, 3-27. Hsu, C.-C. N., Phillips, C., & Yoshida, M. (2005). Cues for Head-final Relative Clauses in Chinese. Poster presented at the 18th Annual CUNY Conference on Human Sentence Processing, University of Arizona, Tucson, AZ. Lin, C.-J. C. (2006). Grammar and Parsing: A Typological Investigation of Relative-Clause Processing. Unpublished Ph.D. Dissertation, The University of Arizona, Tucson. Lin, C.-J. C., & Bever, T. G. (2006). Chinese is no exception: Universal subject preference of relative clause processing. Paper presented at the 19th Annual CUNY Conference on Human Sentence Processing, New York City, NY, March 23–25, 2006. (also available in WCFFL-25 proceedings: http://www.lingref.com/cpp/wccfl/25/index.html) NOTE: Our experiments also contrasted modification types—whether the RCs modified the subject or the object positions of the main clauses. Since it is not relevant to the discussion, we collapsed the data in the presentation. Acknowledgements: This research was supported by an NSC grant (NSC 95-2411-H-003 -056) from Taiwan and an NTNU 2006 research grant to Charles Lin. We thank Larry Li and Li-Hsin Ning for assistance with data collection. We also thank Natalie Hsu for helpful discussions.